Rapid and robust partial least squares regression and its application to NIR spectroscopy analysis

Modern near infrared spectroscopy (NIRS), as an indirect an analytical technique, is used to carry out quantitative analysis of unknown samples by establishing a model with calibration samples. Taking into account the low sensitivity and poor disturbance rejection of NIRS, a new robust version of the SIMPLS algorithm was constructed from a robust covariance matrix for high-dimensional data and robust linear regression in the present paper. Because SIMPLS was based on the empirical cross-covariance matrix between the response variables and the regressors and on linear least squares regression, the results were affected by abnormal observations in the data set. In order to eliminate their negative impact on the accuracy and reliability of the model, a simple multivariate outlier-detection procedure and a robust estimator for the covariance matrix were embedded in the SIMPLS regression framework, based on the use of information obtained from projections onto the directions that maximize and minimize the kurtosis coefficient of the projected data. Finally, application of the proposed kurtosis-SIMPLS method to the NIR analysis was presented with a comparison to the SIMPLS. The results show that kurtosis-SIMPLS method not only finds out the very outliers from the data set with less computational cost, but also holds better prediction performance and steady capability for the normal samples.